rounded rectangle performance in P2D / P3D is extremely poor when stroke is used

MacBook Air M3
Processing 4.5.6 (but also older versions apply)

This runs at around 8.5 fps.

void setup() {
  size(800, 800, P2D);
  frameRate(999);
}

void draw() {
  background(0);

  int cols = 16;
  int rows = 16;
  float margin = 50;
  float gap = 10;

  float rectW = (width - margin * 2 - gap * (cols - 1)) / cols;
  float rectH = (height - margin * 2 - gap * (rows - 1)) / rows;

  fill(255, 220, 0);
  stroke(255);
  strokeWeight(2);
  
  for (int y = 0; y < rows; y++) {
    for (int x = 0; x < cols; x++) {
      float px = margin + x * (rectW + gap);
      float py = margin + y * (rectH + gap);

      rect(px, py, rectW, rectH, 10);
    }
  }

  println(frameRate);
}

首先在 Processing 4.5.6 中运行提供的 Java sketch，并分别在 P2D 和 P3D 中比较启用 stroke 与禁用 stroke 时的帧率。跟踪这些模式使用的圆角矩形渲染路径；当复现不再显示报告中的严重减速，同时保留带 stroke 的圆角矩形时，即表示完成。